# Selecting the right units for large timeline charts

We have very granular timebased data (Captured by the minute) for a very large timeframe (can go into years). This data needs to be depicted on a timeline chart something like this

The green shows normal functioning while the red depicts problem areas. This is depicted over time.

The red areas have some interactivity on hover, however there are certain cases where the time span is set to a year while the data being depicted is in minutes / hours. In such areas the red parts become very tiny, and difficult to hover on.

Basically if there's a problem area that lasted 2 hours, but is being plotted on a day, it works fine, however if the timeline were set to a month or year, the same data shrinks so much that it can't be interacted with or even noticed at times.

I was thinking of solving this by defining specific units based on selected timeline. For e.g. if the timeline selected is a year, we could only depict weeks. So a red chunk would depict a bad week. The values could be calculated as % of smooth function out total time available in the week.

However, I'm a little stuck figuring out what the right units would be and how many use cases can we have given users are free to choose any time range as needed.

So my question is

How does one decide the different units based on selected timeline

Is there a standard scale / table / research in this area

• How frequent are the red chunks? Could they be represented by an "icon" that can be selected?
– Joe
Sep 12, 2016 at 15:38
• The frequency is highly variable, there are cases of errors occurring every 3 hours and lasting for hours as well as every 5 mins and lasting for 3-4 minutes Sep 13, 2016 at 5:33
• I didn't get your point about the icon though Sep 13, 2016 at 6:04

When looking at multiple events over such a long timescale, you're going to run into this kind of problem. Plotting specific events on the "year" scale is feasible - or useful to a user.

You've got to think what the user wants to get out of this information. Looking for a specific event: They're trying to track down the cause of something. They want specific information, and they also know when the event took place. This kind of "yearly" view isn't useful for that.

So what does your user want to get out of this view? It's more likely they want to see a year's worth of events to look for trends in these events. Were they grouped around a particular time? They aren't wanting to know specific event details, but a wider picture.

You're right in that you need to change the metric you present to them. Are they interested in the total downtime? If so, report "Fault Hours". Are they interested in fault numbers? Report the fault frequency. When you break the faults into this kind of metric it makes displaying them easier, as they're all comparable. One good medium would a frequency histogram, either as a classic bar chart or, to fit your current style, a 'heatmap' using a Red Amber Green scale or something. Personally, I find a bar graph:

Much clearer than a 'heatmap':

But either forces you to use some kind of metric you can tally up over any time period. You can just, say, split everything up into arbitrary bins or it might be more helpful to use months/weeks/days. This will show the user if they're getting better or worse. When problematic periods were.

You can also use this to "drill down". Using a classic heat map, you could have three views, Months, Weeks and Day (using your current 'events' view, maybe with icons for small events):

Clicking on the month drills into that month split into weeks, clicking on a week drills down into the week split into days, or something similar.

Ideally you'd have a similar number of "bins" per view and Month-Week-Day doesn't really fit with that, but the user can understand these time periods easily.

Sounds like what you've already got in your head. The important part is coming up with a metric that's understandable, "Number of faults" or "Fault hours" that can easily be explained to the user. I've often had people get confused with frequency histograms - presenting the "bins" as well known periods certainly helps people get their head around them.

• Thanks @Joe, this is exactly why I was thinking of going with weeks or days. I liked your suggestion of using bar charts instead, it is indeed a better representation. However my question was more around the right mapping of these units. Like should the timeline of a year show individual days / weeks or months. I wanted to know if there is some research or set standards / best practices in this area. Sep 13, 2016 at 6:02